Re: [libvirt] [PATCH] lib: implement new API to retrieve the list of CPU models

On 08/21/2013 01:02 PM, Giuseppe Scrivano wrote: Good that the above paragraph is below the ---... ...but this is useful 6 months down the road, and should be in the commit message proper, above the ---. I'm not sure whether returning XML or a straight-up list makes more sense. If you used char ***models, then the user would get an array of directly-usable strings, "486", "pentium", ..., instead of a document that has to be parsed. On the other hand, your idea of returning XML lets us return information for multiple arches simultaneously. But do we need that flexibility, since arch is also an input parameter? Is the idea that you pass arch=NULL to get the full list, or arch="x86" to get the x86 subset of the xml? Why not just make arch mandatory and return char ***; but then you have the question of which arches are supported. So, let's get agreement on the best design before worrying about implementation (I'm still 50/50 on whether xml vs. char*** makes more sense, without more discussion to sway me one way or the other). This is a rather big patch. When adding new API, it's best to do it in pieces: 1. use of the API itself (include, tools, src/driver, src/libvirt*) 2. RPC support (daemon, src/remote) 3. qemu support (src/qemu) 4. test support (src/test) 5. python bindings (python) as long as each part builds and passes 'make check'. I did not review the patch itself, on the grounds that getting the design right, and the patch split, will make review easier. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org